arXiv subjects

Jiyoon Han

Publications and source records attributed to Jiyoon Han.

8 recordsLinked to original sources

arXiv2026

Solar Open 2 Technical Report

We present Solar Open 2, a 250B-A15B Mixture-of-Experts language model built for long-horizon agentic tasks, scaled up from Solar Open 1 (Solar Open 100B). To hold entire agent trajectories in a single context, Solar Open 2 reaches a 1M-token window through a hybrid attention stack that interleaves one softmax layer among every three linear-attention layers, using no positional encoding and a gated delta rule extended to negative eigenvalues. To train at this scale under a fixed compute budget, we make training efficient in two ways: a stronger starting point, and higher-value data. For the starting point, we initialize Solar Open 2 from Solar Open 1, transferring the 5.69B-parameter shared skeleton that survives the architectural change and learning everything else through full pre-training. For the data, we curate for value per token: quality- and rarity-aware data curation and mixture-ratio optimization refine a 20T pool into a 10T mixture that, at equal token budget, outperforms the Solar Open 1 recipe. To build its agent skills, we train twelve domain specialists across purpose-built scenarios, then consolidate them into a single model by Multi-teacher On-Policy Distillation (MOPD). Against comparably sized open-weight models on English benchmarks, Solar Open 2 leads on MMLU-Pro, LiveCodeBench, and the APEX-Agents agentic suite, and stays competitive with the strongest (DeepSeek-V4-Flash and MiMo-V2.5) elsewhere. On Korean benchmarks, Solar Open 2 records the highest average of any model compared, including fast-tier closed APIs, and on Ko-GDPval, an in-house Korean officework-agent benchmark, it is competitive with DeepSeek-V4-Pro (1.6T) at less than a sixth of its size.

cs.CL↗

arXiv2026

MATHENA: Mamba-based Architectural Tooth Hierarchical Estimator and Holistic Evaluation Network for Anatomy

Dental diagnosis from Orthopantomograms (OPGs) requires coordination of tooth detection, caries segmentation (CarSeg), anomaly detection (AD), and dental developmental staging (DDS). We propose Mamba-based Architectural Tooth Hierarchical Estimator and Holistic Evaluation Network for Anatomy (MATHENA), a unified framework leveraging Mamba's linear-complexity State Space Models (SSM) to address all four tasks. MATHENA integrates MATHE, a multi-resolution SSM-driven detector with four-directional Vision State Space (VSS) blocks for O(N) global context modeling, generating per-tooth crops. These crops are processed by HENA, a lightweight Mamba-UNet with a triple-head architecture and Global Context State Token (GCST). In the triple-head architecture, CarSeg is first trained as an upstream task to establish shared representations, which are then frozen and reused for downstream AD fine-tuning and DDS classification via linear probing, enabling stable, efficient learning. We also curate PARTHENON, a benchmark comprising 15,062 annotated instances from ten datasets. MATHENA achieves 93.78% mAP@50 in tooth detection, 90.11% Dice for CarSeg, 88.35% for AD, and 72.40% ACC for DDS.

cs.CV↗

arXiv2026

How Trustworthy Are LLM-as-Judge Ratings for Interpretive Responses? Implications for Qualitative Research Workflows

As qualitative researchers show growing interest in using automated tools to support interpretive analysis, a large language model (LLM) is often introduced into an analytic workflow as is, without systematic evaluation of interpretive quality or comparison across models. This practice leaves model selection largely unexamined despite its potential influence on interpretive outcomes. To address this gap, this study examines whether LLM-as-judge evaluations meaningfully align with human judgments of interpretive quality and can inform model-level decision making. Using 712 conversational excerpts from semi-structured interviews with K-12 mathematics teachers, we generated one-sentence interpretive responses using five widely adopted inference models: Command R+ (Cohere), Gemini 2.5 Pro (Google), GPT-5.1 (OpenAI), Llama 4 Scout-17B Instruct (Meta), and Qwen 3-32B Dense (Alibaba). Automated evaluations were conducted using AWS Bedrock's LLM-as-judge framework across five metrics, and a stratified subset of responses was independently rated by trained human evaluators on interpretive accuracy, nuance preservation, and interpretive coherence. Results show that LLM-as-judge scores capture broad directional trends in human evaluations at the model level but diverge substantially in score magnitude. Among automated metrics, Coherence showed the strongest alignment with aggregated human ratings, whereas Faithfulness and Correctness revealed systematic misalignment at the excerpt level, particularly for non-literal and nuanced interpretations. Safety-related metrics were largely irrelevant to interpretive quality. These findings suggest that LLM-as-judge methods are better suited for screening or eliminating underperforming models than for replacing human judgment, offering practical guidance for systematic comparison and selection of LLMs in qualitative research workflows.

cs.CL↗

arXiv2026

From National Curricula to Cultural Awareness: Constructing Open-Ended Culture-Specific Question Answering Dataset

Large language models (LLMs) achieve strong performance on many tasks, but their progress remains uneven across languages and cultures, often reflecting values latent in English-centric training data. To enable practical cultural alignment, we propose a scalable approach that leverages national social studies curricula as a foundation for culture-aware supervision. We introduce CuCu, an automated multi-agent LLM framework that transforms national textbook curricula into open-ended, culture-specific question-answer pairs for supervised fine-tuning (SFT). Applying CuCu to the Korean national social studies curriculum, we construct KCaQA, comprising 34.1k open-ended QA pairs. Our analyses and training experiments suggest that KCaQA covers culture-specific topics and produces responses grounded in local sociocultural contexts.

cs.CL↗

arXiv2025

Tiled Beamspace MVDR for 1024-element Wideband Radar

We present a tiled architecture for computationally efficient digital beamforming for wideband massive MIMO radar, using beamspace dimension reduction for each tile, and coordinated training of reduced-dimension MVDR beamformers across tiles. We illustrate the efficacy of our approach for a setting in which a 1024-element airborne radar platform beamforms towards airborne targets while suppressing strong interference from ground transmitters. The array is organized into eight 128-element tiles, each a 2D array with 4 (vertical) x 32 (horizontal) elements. Each tile applies a 2D spatial DFT to achieve energy concentration in beamspace, and a 1D temporal FFT to channelize the wideband signal into subbands for which narrowband array models apply. A small tile-level beamspace window is selected for each target (depending on its angle of arrival) in each subband, and coordinated training across tiles is used to compute reduced-dimension MVDR beamformers per-target, per-subband. While full-dimensional MVDR processing is infeasible for the system under consideration, we show that our proposed approach significantly outperforms beamspace MVDR beamforming for a single 128-element tile, where we set the dimensions of the spatial filter (and hence the complexity of MVDR training) to be equal in both systems.

eess.SP↗

arXiv2025

Scaling Wideband Massive MIMO Radar via Beamspace Dimension Reduction

We present an architecture for scaling digital beamforming for wideband massive MIMO radar. Conventional spatial processing becomes computationally prohibitive as array size grows; for example, the computational complexity of MVDR beamforming scales as O(N^3) for an N-element array. In this paper, we show that energy concentration in beamspace provides the basis for drastic complexity reduction, with array scaling governed by the O(NlogN) complexity of the spatial FFT used for beamspace transformation. Specifically, we propose an architecture for windowed beamspace MVDR beamforming, parallelized across targets and subbands, and evaluate its efficacy for beamforming and interference suppression for government-supplied wideband radar data from the DARPA SOAP (Scalable On-Array Processing) program. We demonstrate that our approach achieves detection performance comparable to full-dimensional benchmarks while significantly reducing computational and training overhead, and provide insight into tradeoffs between beamspace window size and FFT resolution in balancing complexity, detection accuracy, and interference suppression.

eess.SP↗

arXiv2022

The Grind for Good Data: Understanding ML Practitioners' Struggles and Aspirations in Making Good Data

We thought data to be simply given, but reality tells otherwise; it is costly, situation-dependent, and muddled with dilemmas, constantly requiring human intervention. The ML community's focus on quality data is increasing in the same vein, as good data is vital for successful ML systems. Nonetheless, few works have investigated the dataset builders and the specifics of what they do and struggle to make good data. In this study, through semi-structured interviews with 19 ML experts, we present what humans actually do and consider in each step of the data construction pipeline. We further organize their struggles under three themes: 1) trade-offs from real-world constraints; 2) harmonizing assorted data workers for consistency; 3) the necessity of human intuition and tacit knowledge for processing data. Finally, we discuss why such struggles are inevitable for good data and what practitioners aspire, toward providing systematic support for data works.

cs.HC↗

arXiv2021

KLUE: Korean Language Understanding Evaluation

We introduce Korean Language Understanding Evaluation (KLUE) benchmark. KLUE is a collection of 8 Korean natural language understanding (NLU) tasks, including Topic Classification, SemanticTextual Similarity, Natural Language Inference, Named Entity Recognition, Relation Extraction, Dependency Parsing, Machine Reading Comprehension, and Dialogue State Tracking. We build all of the tasks from scratch from diverse source corpora while respecting copyrights, to ensure accessibility for anyone without any restrictions. With ethical considerations in mind, we carefully design annotation protocols. Along with the benchmark tasks and data, we provide suitable evaluation metrics and fine-tuning recipes for pretrained language models for each task. We furthermore release the pretrained language models (PLM), KLUE-BERT and KLUE-RoBERTa, to help reproducing baseline models on KLUE and thereby facilitate future research. We make a few interesting observations from the preliminary experiments using the proposed KLUE benchmark suite, already demonstrating the usefulness of this new benchmark suite. First, we find KLUE-RoBERTa-large outperforms other baselines, including multilingual PLMs and existing open-source Korean PLMs. Second, we see minimal degradation in performance even when we replace personally identifiable information from the pretraining corpus, suggesting that privacy and NLU capability are not at odds with each other. Lastly, we find that using BPE tokenization in combination with morpheme-level pre-tokenization is effective in tasks involving morpheme-level tagging, detection and generation. In addition to accelerating Korean NLP research, our comprehensive documentation on creating KLUE will facilitate creating similar resources for other languages in the future. KLUE is available at https://klue-benchmark.com.

cs.CL↗